Doubly robust nonparametric instrumental variable estimators for survival outcomes.

Area of Science:

• Biostatistics
• Causal Inference
• Survival Analysis

Background:

• Unmeasured confounding is a major challenge in causal inference.
• Existing instrumental variable (IV) methods are limited for censored survival outcomes.
• Developing robust methods for survival data is crucial.

Purpose of the Study:

• To propose novel nonparametric estimators for local average treatment effect on survival probabilities.
• To address both covariate-dependent and outcome-dependent censoring.
• To provide efficient and flexible IV methods for survival data.

Main Methods:

• Developed efficient influence function-based estimators.
• Proposed simple estimation procedures for binary or continuous instrumental variables.
• Incorporated machine learning tools for nonparametric estimation.
• Ensured double-robustness properties of the estimators.

Main Results:

• Demonstrated flexibility and double robustness in simulation studies under various scenarios.
• Successfully applied the method to the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial.
• Estimated the causal effect of screening on survival probabilities.
• Investigated causal contrasts under different censoring assumptions.

Conclusions:

• The proposed nonparametric IV estimators effectively address unmeasured confounding in censored survival outcomes.
• The methods offer flexibility and double robustness, enhancing causal inference in survival analysis.
• Applicable to real-world health studies for estimating treatment effects on survival.